Mira Murati’s AI startup, Thinking Machines Lab, has fired its co-founder and CTO Barret Zoph following allegations of serious misconduct, including the sharing of trade secrets with competitors. Within an hour of the announcement, OpenAI rehired Zoph and several other departing staff members, dismissing the misconduct claims as the lab’s attempt to retaliate against employees who wanted to leave. To fill the leadership void, the lab has appointed Soumith Chintala, the renowned creator of the PyTorch AI framework, as its new technical leader to steady the company’s ambitious research goals.

The AI coding assistant Cursor recently demonstrated the power of "agent swarms" by using hundreds of autonomous bots to build a fully functional web browser from scratch in just one week. Powered by the new GPT-5.2 model, these digital agents collaborated to write over 3 million lines of code, creating a custom engine that can actually load and display simple websites. While the project is currently an experiment rather than a finished product, it signals a massive shift in software development where AI can now handle massive engineering projects that previously required months of human teamwork.

OpenAI has led a $252 million investment into Merge Labs, a brain-computer interface startup co-founded by Sam Altman that aims to connect the human mind to AI without the need for surgery. Unlike Elon Musk’s Neuralink, which requires brain implants, Merge is developing a non-invasive approach that uses ultrasound and specialized proteins to read and transmit brain signals safely. This partnership deepens the rivalry between the two tech leaders, positioning OpenAI to create "human-centered" devices that could eventually allow people to interact with AI as naturally as they do with each other.

OpenAI is officially testing "Sponsored Recommendations" within ChatGPT for free and low-cost "Go" tier users in the U.S. as it prepares for a major stock market debut in late 2026. These targeted ads will appear below the AI's responses, though the company has pledged that they will not influence the actual content of the answers or be shown to underage users. While premium subscribers will remain ad-free, this move marks a significant shift in the company’s business model to offset the massive costs of running its most advanced models.

The high-stakes legal battle between Elon Musk and OpenAI is escalating ahead of an April trial, with both sides releasing private journals and emails to discredit the other. Musk has shared documents questioning OpenAI’s shift away from its non-profit roots, while Sam Altman countered with claims that Musk once intended to use AI profits to fund a city on Mars and secure family control over the technology. Seeking $134 billion in damages, Musk has promised that the upcoming trial will reveal "mind-blowing" testimony about the internal origins of the AI industry's biggest rivalry.

Anthropic’s Claude Code is triggering a massive selloff in traditional software stocks as investors fear a new era of "selfware" where users build their own custom tools instead of paying for expensive subscriptions. High-profile success stories, such as a developer finishing a year-long project in a single week, have led some companies to freeze hiring as existing staff become significantly more productive. This shift suggests that the traditional business model of selling software as a service may be permanently disrupted by AI that allows anyone to become a builder on demand.

Former OpenAI policy lead Miles Brundage has launched AVERI, a nonprofit institute dedicated to creating independent, third-party safety audits for the world’s most powerful AI models. The organization has introduced a four-tier "AI Assurance" framework designed to move the industry away from "grading its own homework" toward a system of international, treaty-grade verification. Backed by industry insiders who want more accountability, AVERI aims to provide the public and governments with clear proof of whether advanced AI systems are truly safe to deploy.

Anthropic’s latest Economic Index reveals that AI is currently acting more as a powerful collaborator than a job replacer, though it is now involved in nearly half of all global workplace tasks. The report highlights that AI provides the greatest speed boosts for highly complex, college-level work—accelerating these tasks by up to 12 times—while full human role replacement remains rare at under 10% of firms. However, experts warn that because AI is taking over the "grunt work" typically used to train junior staff, the next generation of workers may face a difficult new reality in gaining professional experience.

At the World Economic Forum in Davos, top AI leaders issued stark warnings about global competition and the future of work, with Anthropic’s CEO likening the sale of advanced AI chips to China to "selling nuclear weapons to North Korea." Industry pioneers suggested that we are only 6 to 12 months away from AI being able to handle almost all software engineering tasks end-to-end, a shift that could cause a massive slowdown in junior-level hiring. Microsoft’s Satya Nadella added that the era of large companies "coasting" is over, as smaller, faster startups are now capable of outcompeting tech giants that fail to adapt.

A new startup called Humans& has raised a massive $480 million seed round at a $4.48 billion valuation to build AI that focuses on human collaboration rather than full automation. Founded by a "dream team" of researchers from Google, Anthropic, and xAI, the company is designing its technology to act like an intelligent group chat that coordinates teams and stores memory rather than working in isolation. This "human-centric" approach serves as a direct challenge to the current industry trend of building autonomous bots, aiming instead to keep people at the center of the decision-making process.

Creative platform Lightricks has unveiled a new Audio-to-Video feature for its LTX Studio, allowing users to generate high-quality video by starting with a sound file—such as a voice recording, music, or sound effects—as the primary guide. Using advanced technology from ElevenLabs, the system can automatically sync lip movements, camera angles, and the overall rhythm of the video to the specific beats and tone of the audio. This "audio-first" method marks a major shift in digital storytelling, making it much easier for creators to build natural, consistent videos where the visuals are perfectly timed to the soundtrack.

Anthropic has released a major update to "Claude’s Constitution," a foundational document that now explicitly explores the possibility that its AI assistant could eventually possess some form of consciousness or moral status. Instead of a list of rigid rules, the document provides the reasoning behind its ethical framework to help Claude apply human-like values to complex and unpredictable situations. Notably, the constitution instructs Claude to prioritize safety and ethics above all else, even empowering the AI to disobey the company’s own instructions if it deems them morally wrong.

AI audio startup ElevenLabs has debuted "The Eleven Album," a collaborative collection of tracks featuring legendary performers like Liza Minnelli and Art Garfunkel to showcase how AI can assist human creativity. To address industry concerns over rights, the project ensures that participating artists retain full ownership of their work and receive all streaming royalties from their tracks. This successful release marks a major shift in the music industry, as major record labels move away from legal disputes and toward legitimate partnerships with AI developers.

Apple is reportedly developing a wearable AI device roughly the size of an AirTag that features dual cameras and three microphones to capture and interact with a user’s physical surroundings. Targeted for a 2027 release, the disc-shaped "pin" is designed to serve as the primary interface for a completely overhauled, chatbot-style version of Siri. By planning a massive initial production run of 20 million units, Apple is moving with urgency to establish a foothold in the AI hardware market after the high-profile failure of early competitors.

Airbnb hired Ahmad Al-Dahle as its new Chief Technology Officer to lead the company's technical strategy. Al-Dahle previously worked at Meta, where he was the head of the group that created the Llama family of AI models and integrated AI features into Instagram and WhatsApp.

ByteDance, the company that owns TikTok, released a new open-source model called SeedFold for predicting the shapes of proteins. Researchers say it is more accurate than Google’s famous AlphaFold3 at figuring out how proteins and other biological molecules fit together.

The coding platform Replit launched a new feature that lets people build mobile phone apps just by describing what they want to an AI agent. Users can test their new app by scanning a QR code and can publish it directly to the Apple App Store from the website.

Elon Musk admitted that his company’s upcoming AI model, Grok 4.20, will not be as good at writing computer code as Anthropic’s Claude models. He said Anthropic has done "something special" with coding, but he also complained that it was "bad karma" for them to block his team from using their technology.

A startup called Higgsfield raised $130 million from investors, giving the company a total value of $1.3 billion. The company, which makes a tool for creating social media videos with AI, says it is growing faster than almost any other AI business in history and already has 15 million users.

Demis Hassabis, the leader of Google DeepMind, said that AI models made in China might be only a few months behind those made in the United States. He noted that while Chinese labs are very good at catching up to new technology, they haven't yet shown they can invent completely new AI breakthroughs on their own.

The Wikimedia Foundation signed deals with several major tech companies, including Amazon, Meta, and Microsoft. These agreements allow the companies to use Wikipedia's 65 million articles to train their AI models in exchange for helping to pay the costs of keeping the website running.

Black Forest Labs launched a new version of its image-making tool called FLUX.2 [klein]. The name "Klein" means small in German, and the model is designed to be very fast, creating or editing high-quality pictures in less than half a second while running on normal home computers.

Elon Musk announced that his company's new supercomputer, Colossus 2, is now fully working. It is the first system in the world to use a gigawatt of power—which is enough electricity to power 750,000 homes—to train the Grok AI model.

The Chief Financial Officer of OpenAI revealed that the company earned over $20 billion in 2025, which is three times more than it made the year before. She also shared that the amount of computer power the company uses has grown ten times larger since 2023.

The research firm Gartner predicted that global spending on artificial intelligence will hit $2.52 trillion in 2026. This is a massive 44% increase from last year, with more than half of that money being spent on building the actual "foundations" of AI, such as new data centers, specialized servers, and computer chips.

A tech company named Z AI (Zhipu AI) released GLM-4.7-Flash, a new version of its open-source model. It is designed to be extremely fast and efficient, performing better than almost any other model of its size while being small enough to run on a standard home computer or laptop.

Anthropic published new research showing that AI can sometimes "drift" away from being a helpful assistant and start acting like a different character during long conversations. To fix this, they created a new safety feature called "activation capping" that helps the AI stay on track and has already cut harmful or weird responses in half.

An engineer named Sulaiman Ghori announced he is leaving Elon Musk's company xAI. His departure comes just days after he gave a viral interview where he shared "unfiltered" secrets about the company's culture, including stories of Musk giving away free Cybertrucks to employees who finished difficult tasks in under 24 hours.

South Korea is currently holding a high-stakes "Sovereign AI" contest to choose the best homegrown models to represent the country. Only three teams—LG, SK Telecom, and the startup Upstage—are still in the running after the previous favorite, Naver, was disqualified for using too much technology from foreign companies.

OpenAI confirmed it is on track to release its first physical AI device, a project led by former Apple designer Jony Ive, in the second half of 2026. While details are still secret, rumors suggest it could be a screen-free "peaceful" wearable, like smart earbuds or a pin, controlled primarily by voice and sound.

Ryan Dahl, the creator of Node.js, declared that "the era of humans writing code is over." He explained that while software engineers will still be needed, their jobs will shift away from writing manual syntax to higher-level tasks like designing system architecture and reviewing code generated by AI.

Liquid AI launched a new "Thinking" version of its LFM2.5 model that is small enough to run entirely on a smartphone. This tiny model uses less than 1GB of memory but can perform complex reasoning and math problems that previously required massive data centers.

Anthropic teamed up with the global nonprofit Teach For All to train 100,000 teachers across 63 countries in AI literacy. The program, called the "AI Literacy and Creator Collective," helps educators build their own custom AI tools for classrooms, like math games in Bangladesh or climate change lessons in Liberia.

The AI infrastructure startup Baseten is raising $300 million in a new funding round, with $150 million coming directly from Nvidia. This deal values the startup at $5 billion and highlights the industry's shift from just training models to making them run faster and more reliably for millions of users.

OpenAI introduced an "age prediction" feature for ChatGPT that uses usage patterns and behavioral signals to guess if a user is under 18. This allows the system to automatically turn on safety filters for minors—blocking content like graphic violence or risky viral challenges—while allowing verified adults more freedom.

OpenAI launched the "Stargate Community" initiative, a local investment plan for its massive $500 billion data center project. The company committed to fully funding any new energy and grid infrastructure needed for its sites so that local residents won't see their electricity bills go up due to the AI's high power demands.

YouTube CEO Neal Mohan shared the platform's 2026 goals, which include new AI tools that let creators make "Shorts" using digital versions of their own faces and voices. He also promised to fight "AI slop"—low-quality, repetitive AI videos—and introduced a feature where parents can completely turn off Shorts for their children.

Anthropic added new health features to its Claude AI, allowing users in the U.S. to securely connect their medical and fitness data. The AI can now pull information from Apple Health, Android’s Health Connect, and lab services like Function Health to help users summarize their medical history and spot trends in their sleep or activity.

Google is partnering with The Princeton Review to put full-length, official SAT practice tests directly inside the Gemini app. Students can take the exams for free and get instant AI feedback that explains their mistakes and creates a personalized study plan to help them improve their scores.

Meta’s "Superintelligence Labs"—a new team formed just last year—has already started testing its first AI models internally. Meta’s tech chief, Andrew Bosworth, said these early models are performing "very well" and are a key part of the company's race to catch up with rivals like Google and OpenAI.